Spatio-temporal analysis for automatic contrast injection detection on angiography during trans-catheter aortic valve implantation

Abstract

A method that includes generating a contrast feature curve for a medical image sequence including a plurality of frames, where the contrast feature curve represents contrast feature values of the frames. The method further includes detecting a peak in the contrast feature curve, and determining whether the peak corresponds to at least one of contrast injection in an aortic root, contrast injection in a balloon, and a non-contrast injected region.

What is claimed is: 
     
       1. A method, comprising:
 generating a contrast feature curve for a medical image sequence comprising a plurality of frames, wherein the contrast feature curve represents contrast feature values of the frames; 
 detecting a peak in the contrast feature curve; and 
 determining whether the peak corresponds to at least one of contrast injection in an aortic root, contrast injection in a balloon, and a non-contrast injected region; 
 wherein when the contrast feature values of the peak are less than a first threshold and greater than a second threshold, the method further comprises: 
 extracting features from the medical image sequence and feeding the features into a support vector machine (SVM) to train the SVM; and 
 classifying frames in the neighborhood of the peak as being of the contrast injected aorta or not having contrast injection based on the number of neighboring frames classified as contrasted using the SVM. 
 
     
     
       2. The method of  claim 1 , wherein the medical image sequence is fluoroscopic or angiographic. 
     
     
       3. The method of  claim 1 , wherein a contrast feature value of a frame is a ratio of histogram similarity at that frame to a first reference histogram representing the aortic root with contrast injection and a second reference histogram representing the aortic root without contrast injection. 
     
     
       4. The method of  claim 1 , wherein when the contrast feature values of the peak are greater than a first threshold, the method further comprises:
 generating a spatio-temporal feature map which identifies an arrival time of contrast injection at each pixel; 
 identifying, in the spatio-temporal feature map, a distribution pattern of the time-of-arrival; and 
 determining that the spatio-temporal feature map is of the aortic root if the time-of-arrival is smallest near a lower edge of the spatio-temporal feature map and increases upward, or that the spatio-temporal feature map is of the balloon if the time-of-arrival is smallest near a center of the spatio-temporal feature map and increases outward. 
 
     
     
       5. The method of  claim 4 , wherein a shape similarity score is calculated using shape matching to a segmentation of the spatio-temporal feature map to further determine that the medical image sequence is of the aortic root or the balloon. 
     
     
       6. The method of  claim 4 , wherein a cascaded classifier is used to combine the distribution pattern of the time-of-arrival and a shape-similarity score to detect a balloon. 
     
     
       7. The method of  claim 4 , further comprising:
 determining which frames neighboring the frames of the peak have contrast injection, wherein this determination is performed using a classifier trained using the frames of the peak known to have contrast injection and frames not neighboring the peak known not to have contrast injection. 
 
     
     
       8. The method of  claim 1 , wherein when the contrast feature values of the peak are less than a second threshold the peak corresponds to the non-contrast injected region. 
     
     
       9. A system, comprising:
 a memory device for storing a program;
 a processor in communication with the memory device, the processor operative with the program stored on a non-transitory computer readable medium to: 
 generate a contrast feature curve for a medical image sequence comprising a plurality of frames, wherein the contrast feature curve represents contrast feature values of the frames; 
 detect a peak in the contrast feature curve; and 
 determine whether the peak corresponds to at least one of contrast injection in an aortic root, contrast injection in a balloon, and a non-contrast injected region; 
 wherein when the contrast feature values of the peak are less than a first threshold and greater than a second threshold, the processor is operative with the program to: 
 extract features from the medical image sequence and feed the features into a support vector machine (SVM) to train the SVM; and 
 classify frames in the neighborhood of the peak as being of the contrast injected aorta or not having contrast injection based on the number of neighboring frames classified as contrasted using the SVM. 
 
 
     
     
       10. The system of  claim 9 , wherein the medical image sequence is fluoroscopic or angiographic. 
     
     
       11. The system of  claim 9 , wherein a contrast feature value of a frame is a ratio of histogram similarity at that frame to a first reference histogram representing the aortic root with contrast injection and a second reference histogram representing the aortic root without contrast injection. 
     
     
       12. The system of  claim 9 , wherein when the contrast feature values of the peak are greater than a first threshold, the processor is operative with the program to:
 generate a spatio-temporal feature map which identifies an arrival time of contrast injection at each pixel; 
 identify, in the spatio-temporal feature map, a distribution pattern of the time-of-arrival; and 
 determine that the spatio-temporal feature map is of the aortic root if the time-of-arrival is smallest near a lower edge of the spatio-temporal feature map and increases upward, or that the spatio-temporal feature map is of the balloon if the time-of-arrival is smallest near a center of the spatio-temporal feature map and increases outward. 
 
     
     
       13. The system of  claim 12 , wherein a shape similarity score is calculated using shape matching to a segmentation of the spatio-temporal feature map to further determine that the sequence is of the aortic root or the balloon. 
     
     
       14. The system of  claim 12 , wherein a cascaded classifier is used to combine the distribution pattern of the time-of-arrival and a shape-similarity score to detect a balloon. 
     
     
       15. The system of  claim 12 , wherein the processor is operative with the program to:
 determine which frames neighboring the frames of the peak have contrast injection, wherein this determination is performed using a classifier trained using the frames of the peak known to have contrast injection and frames not neighboring the peak known not to have contrast injection. 
 
     
     
       16. The system of  claim 9 , wherein when the contrast feature values of the peak are less than a second threshold the peak corresponds to the non-contrast injected region. 
     
     
       17. A computer program product, comprising:
 a non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code comprising: 
 computer readable program code configured to perform the steps of: 
 generating a contrast feature curve for a medical image sequence comprising a 
 plurality of frames, wherein the contrast feature curve represents contrast feature values of the frames; 
 detecting a peak in the contrast feature curve; and 
 determining whether the peak corresponds to at least one of contrast injection in an aortic root, contrast injection in a balloon, and a non-contrast injected region; 
 wherein when the contrast feature values of the peak are less than a first threshold and greater than a second threshold, the processor is operative with the program to: 
 extract features from the medical image sequence and feed the features into a support vector machine (SVM) to train the SVM; and 
 classify frames in the neighborhood of the peak as being of the contrast injected aorta or not having contrast injection based on the number of neighboring frames classified as contrasted using the SVM.